Hector: Using Untrusted Browsers to Provision Web Applications

TL;DR

Hector is a novel web application framework that leverages untrusted browsers for distributed computing, improving latency and reducing costs by offloading logic to clients while ensuring security and scalability.

Contribution

Hector introduces a new framework that transforms browsers into a distributed, application-centric platform using WebAssembly, WebRTC, and trusted execution, enabling scalable and cost-effective web applications.

Findings

Achieves lower end-user latency compared to traditional deployments.

Reduces operational costs for service providers.

Scales linearly with increasing client numbers and handles unresponsive clients effectively.

Abstract

Web applications are on the rise and rapidly evolve into more and more mature replacements for their native counterparts. This disruptive trend is mainly driven by the attainment of platform-independence and instant deployability. On top of this, web browsers offer the opportunity for seamless browser-to-browser communication for distributed interaction. In this paper, we present Hector, a novel web application framework that transforms web browsers into a distributed application-centric computing platform. Hector enables offloading application logic to users, thereby improving user experience with lower latencies while generating less costs for service providers. Following the programming paradigm of Function-as-a-Service, applications are decomposed into functions so they can be managed efficiently and deployed in a responsive, scalable and lightweight fashion. In case of client-side resource shortage or unresponsive clients, execution falls back to a traditional cloud-based infrastructure. Hector combines WebAssembly for multi-language computations at near-native speed, WebRTC for browser-to-browser communication and trusted execution as provided by the Intel Software Guard Extensions so browsers can trust each other's computations. We evaluate Hector by implementing a digital assistant as well as a recommendation system. Our evaluation shows that Hector achieves lower end-user latencies while generating less costs than traditional deployments. Additionally, we show that Hector scales linearly with increasing client numbers and can cope well with unresponsive clients.